On January 21, a photo of an Nvidia GPU with code name GB202-200-A1 (see header image above) surfaced on the infamous Chip Hell forum. The user who posted it claims that it’s a prototype of the most powerful GPU in the upcoming RTX 50 series.
The fully configured GPU allegedly utilizes the maximum possible number of computing units for which the chip was designed. We’re talking about an RTX 50-series chip that supports 192 streaming processors and an incredible 24,576 shader units. The current top model RTX 5090 has “only” 21,760 shader units. Mathematically, the GB202-200-A1 has almost 13 percent more raw computing power.
The second striking feature are the two 12V 2×6 power connections. These also appear to be necessary, as the Thermal Design Power (TDP) is said to be an extremely high 800 watts. By comparison, the RTX 5090 has a TDP of 575 watts. The power consumption of the GB202-200-A1 is therefore greater by 39 percent.
| GB202-200-A1 | GeForce RTX 5090 | |
| GPU | GB202-200-A1 | GB202-300-A1 |
| Codename | David Blackwell | David Blackwell |
| Computing power (FP32) | 123.6 TFLOPS | 104.9 TFLOPS |
| Shader units | 24,576 | 21,760 |
| Streaming processors | 192 | 170 |
| GPU clock base / boost | 2,100MHz / 2,514MHz | 2,010MHz / 2,410MHz |
| Memory bus | 512 bit | 512 bit |
| Memory type | GDDR7 | GDDR7 |
| Graphics memory | 32GB | 32GB |
| TGP | 800 watts | 575 watts |
| RRP | ? | $1,999 |
Nvidia appears to be investing that additional power requirement in higher clock frequencies. Here, too, the Chip Hell forum thread provides very specific figures: the base and boost clocks of the GB202-200-A1 are said to be 2,100MHz and 2,514MHz, respectively. For comparison, the RTX 5090 runs at 2,010MHz and 2,410MHz.
As with the RTX 5090, the RAM configuration should be 32GB, which are connected via a 512-bit memory interface. The picture clearly shows the 16 memory chips surrounding the graphics processor. But here, too, Nvidia could go one better — by only installing chips with a capacity of 3GB, as is already the case with the mobile version of the RTX 5090.
Now the only question that remains is what it should be called. It would be conceivable to bring the GB202-200-A1 to market as the RTX 5090 Ti. But a new entry in the Titan series would also be possible, perhaps as the Nvidia Titan RTX Blackwell. Since the Titan is traditionally aimed more at consumers in the creative sector (e.g., content creators), I’d be disappointed if this card didn’t have 48GB of GDDR7 memory.